Most needed AI Change is...?

[thadian]

There are 3 changes i feel need to be made to the AI. Each of these i have either read about or experienced in the game that i felt made bad gameplay.

Unpromoted Units: The AI should wait to promote its units until it knows what its enemy is using. When its ready to attack, it should scan the "battlefield" and promote against the unit types it will be up against.

Smarter AI combat: The AI should build stacks of units based on what it needs to win - it shouldn't build all horse archers because its the only one with horses for example. It should build responsively to the other units being built around it.

AI Patrol: AI Should use patrol for both land and sea units, which should provide it more internal security instead of "hiding" in its cities.

 

[EweezE]

Those sound good. I have one to add to the list, rather than starting my own thread..

AI Trading Blunder: When you've invested enough EP in an AI to see their research, you can exploit the AI. Wait until they are 2 turns away from discovering a Tech you already have and then sell it to them for ~100 Gold. I can't resist doing this and feel guilty sometimes after doing so. Maybe this can be fixed somehow? I play with World Builder disabled but there's nothing to save me from this exploit except disabling technology trading. Tech trading is a neat feature which adds diplomacy, so it's nice to leave it enabled. I'm only human, and humans exploit the hell out of stuff.. I think this needs a fix as well thadian. What do you think?

 

[thadian]

i have noticed that and agree - even worse is when its your vassal and you switch their research until almost all of their techs are 2-3 turns off and take them for all theyre worth.

Theres also some articles on an exploit about how to subsidize them in trading. its simple to do.

find an AI with 8 GPT for trade.

give them 1GPT until it stops going up. lets say you get to 12.

Sell them excess wheat for 12 gold.

Gift them 12 gold

Sell them excess spice for 12 gold

Gift them 12 gold

once the time limit expires - just cancel the "gifts" and you stick them with a pretty nice debt, in addition to pawning off useless resources at the high price. last night when i made this i subsidized the whole table for 200 gpt running 100% research!

I have seen occasional "cutoff" periods for AI where they don't trade for a tech they are 2 turns from researching - but that is rare. I also noticed when you can see everything they build, you can also move spies into their major city. when they are done with apolo program, use your spies to ruin their project. all that time he sat there doing nothing can make this a very cheap way to cheat the AI a space-race victory by never allowing them to complete it.

 

[DanF5771]

AI Trading Blunder: When you've invested enough EP in an AI to see their research, you can exploit the AI. Wait until they are 2 turns away from discovering a Tech you already have and then sell it to them for ~100 Gold. I can't resist doing this and feel guilty sometimes after doing so. Maybe this can be fixed somehow? I play with World Builder disabled but there's nothing to save me from this exploit except disabling technology trading. Tech trading is a neat feature which adds diplomacy, so it's nice to leave it enabled. I'm only human, and humans exploit the hell out of stuff.. I think this needs a fix as well thadian. What do you think?

Play with "No Tech Brokering" on, this will not allow you to sell techs, the AI has already invested >50% of the necessary beakers in. But you can still trade techs.

--

Option 3 is good as long as the Patrol functions are implemented to work properly. As we all know, this is currently not the case for Sea Patrol.

 

[PimpyMicPimp]

I voted option one, since if the AI a) Used City Raider Promotions and B) had a better grasp on using units in general, it would be far tougher to defeat it in war, which is it's weakest area by far. We've already seen amazing strides from Vannilla AI to BTS AI, but it's still fairly weak in combat.

 

[MrCynical]

One area which would be relatively easy to fix is the AI tech trading, which is almost deliberately incompetant for most civs.

 

[IronCrown]

AI Trading Blunder: When you've invested enough EP in an AI to see their research, you can exploit the AI. Wait until they are 2 turns away from discovering a Tech you already have and then sell it to them for ~100 Gold.

What's wrong with that? The AI saves 2 turns research time and pays 100 gold for that. Where's the exploit?

 

[r_rolo1]

AI tactical manouvering ... AI armies move in very predictable ways and it is easy to hole them in traps or to divert them out of the real combat area...

 

[Supr49er]

The AIs like to launch amphibious attacks with units that don't have Amphibious promotions instead of landing their stack and using it correctly.

 

[Magma_Dragoon]

The AI trading gold for a tech it almost has isn't an exploit. It will pay an amount of gold based on how many beakers are left unresearched. If you do it with a very early tech, you won't get very much, but the later techs cost much more and the AI will pay much more.

 

[EweezE]

Play with "No Tech Brokering" on, this will not allow you to sell techs, the AI has already invested >50% of the necessary beakers in. But you can still trade techs.

Are you sure about that Dan? I thought Tech Brokering only disallows you to sell/trade techs that you did not discover yourself. So for example, if I research Literature myself then I can still do the exploit. However, if I acquired Literature through trade then I can not turn around and trade it to someone else. I haven't played with No Tech Brokering yet, but that is my understanding from the way it reads.

What's wrong with that? The AI saves 2 turns research time and pays 100 gold for that. Where's the exploit?

It's highway robbery! Would you pay an AI 100 gold for a tech you were 2 turns away from? The AI makes these kind of trades in the early game when your gold per turn is real low due to expansion. If it takes an AI ~30 turns to generate 100 gold and then you rip them like that.. You can really disrupt their research slider. Force them to tax more to up their treasury and then do it over again once they are at 2 turns on the next tech. It's a cruel cycle which benefits the player because he can basically run in a defecit if he wishes (favoring research), knowing he's got ~100 gold waiting for him in X turns through trade thanks to his espionage intel. An AI should refuse trade for a tech once they have invested >50% of the necessary beakers in it as Dan mentioned.

 

[IronCrown]

I don't think so. How much gold an AI pays for a tech is subject to the same rules that apply to a tech-tech trade. If an AI offers me 100 gold for a tech that is indeed usually highway-robbery - but not on my part

 

[DanF5771]

Are you sure about that Dan? I thought Tech Brokering only disallows you to sell/trade techs that you did not discover yourself. So for example, if I research Literature myself then I can still do the exploit. However, if I acquired Literature through trade then I can not turn around and trade it to someone else. I haven't played with No Tech Brokering yet, but that is my understanding from the way it reads.

Yes, I'm pretty sure. Kurtkage started a thread about this in Solver's unofficial patch sub-sub-forum. The AI will refuse a tech (even as a gift) with "We would have nothing to gain". This makes perfectly sense, because by selling (gifting) it to the AI, you would deny them any further trades = exploit. So whenever their self-research progress is >50% they won't accept it.

 

[r_rolo1]

Would you pay an AI 100 gold for a tech you were 2 turns away from?

if i lacked more than 100 for finishing the tech, probably yes.

 

[DanF5771]

If I have an empire that is highly specialized for a max slider (90/100%) with a lot of beaker multiplier buildings (libraries, monasteries, universities, academies), then these deals are very much to my advantage.

 

[EweezE]

If an AI offers me 100 gold for a tech that is indeed usually highway-robbery - but not on my part

I see you agree with it being highway robbery... but who do you think is getting robbed? "but not on my part" doesn't make any sense. Does that mean that you just got lucky and they offered you ~100 gold for the tech? Is it still just getting lucky when you get the gold for almost all techs they research? Am I supposed to pat myself on the back for my Tech Trading skills, or do you now see that this is an exploit?

Unless there is some insane post, I'm done discussing the exploit thadian .

 

[thadian]

I think the AI should have a "War Goal" such as "Capture 4 cities" or "Capture capital". Most players go into war with said goals, right? It should notice when you are trying to hole up its attack force, or using pawns to buy a turn, or divert it away from your key cities.

One thing i learned in multi-player is to stagger your wars into 3 stages. Build the borders, sack the borders, go to peace and rebuild your scattered troops and war again when your stacks are rebuilt (capturing 3 cities is good enough for me in a war). The AI should do similar.

I know i can not ask for a "humanlike" AI, but i should at least get one that knows not just the system its playing under, but one who knows how to win.

I think it would be nice if each AI would have its own victory condition it aims for. (Ceasar and Boadica want to end the game early, because they dont want gaul swordsman and pretorians staring down cannons and infantry).

Likewise, Ghandi wants to aim for a cultural/diplo victory when possible.



So far it seems that most of us agree the main AI flaw is that it does not know how to manage its units according to what it's enemies are using, weather it is in how the units are promoted / what is built to begin with - and with the techniques they use to move their armies.

 

[Zeestein]

I'm really more of a builder than a warmonger, but when I war I war with strategic reasons or with the intent to KILL off a weaker civ. Last game I had a decent tech lead and killed off my only rivals on the continent, and I fought 2 defensive amphibious wars in the renaissance era vs about 20 units each time, coming in separate galleon drops, about 1/2 trebs and 1/2 cuirassers landing to capture weak cities, and I mean weak, as in 3 and 2 pop newly settled cities. I mean, they were rivals but their attempt at conquest was a joke.

So it's this: strategic planning in warfare by the AI and better mechanics RE amphibious attacks.

 

[The Snug]

There are 3 changes i feel need to be made to the AI. Each of these i have either read about or experienced in the game that i felt made bad gameplay.

Unpromoted Units: The AI should wait to promote its units until it knows what its enemy is using. When its ready to attack, it should scan the "battlefield" and promote against the unit types it will be up against.

Smarter AI combat: The AI should build stacks of units based on what it needs to win - it shouldn't build all horse archers because its the only one with horses for example. It should build responsively to the other units being built around it.

AI Patrol: AI Should use patrol for both land and sea units, which should provide it more internal security instead of "hiding" in its cities.

For 1 and 2, the AI does respond appropriately on Marathon level for long, protracted wars after it loses its initial stack. On marathon, however, if the human notices the AI adapting, the human can either sue for peace, or also counter-the-counter. It's probably less noticeable on slower game speeds when units obsolete so fast that adaptation means upgrading to the next era of units.

For 3, if the function actually worked probably, I'd really go for this because it would make attacking much more difficult. As it stands now, however, if the AI actually implemented the current Patrol feature on both land and sea, then I would both slaughter (can you say, guerrilla!) the AI to no end and amass GG's in the process. But yes, the AI needs better defensive measures, bcz has anyone ever seen the AI defend itself successfully?

 

[TheMeInTeam]

AI tactical manouvering ... AI armies move in very predictable ways and it is easy to hole them in traps or to divert them out of the real combat area...

Yeah right.

I agree with you, but it's a pipe dream for now . Any coding other than just attacking in logical pathing places that is consistent would allow the human player to exploit it. I guess you could just have the AI randomly do things, like send more pillage stacks and not a SoD, or split SoD, or naval invasion etc., but without great care it could cause a lot of AI stupidity - and more of that than we're used to would be too much .

Maybe the best way would be to hard code like 6-8 attack/defense patterns patterns, and have the AI hit with them at random. The player might know that the AI will attack X city with a naval invasion, and counter it. HOWEVER, if rather than "will attack" it becomes "has a chance to attack there 1/4 of the time, but also may hit y, z, etc", it would place considerably more duress on the player in terms of dealing with the AI. I think this approach should be used in general AI tactics - hardcoding things that would be exploitable, except that the AI has a chance to do other things that going for the exploit would be weaker against.

Of course, the AI might have to lose some of its bonus power if that's done effectively .